Multimodal evaluation of hypoxia in brain metastases of lung cancer and interest of hypoxia image-guided radiotherapy

Lung cancer patients frequently develop brain metastases (BM). Despite aggressive treatment including neurosurgery and external-radiotherapy, overall survival remains poor. There is a pressing need to further characterize factors in the microenvironment of BM that may confer resistance to radiotherapy (RT), such as hypoxia. Here, hypoxia was first evaluated in 28 biopsies from patients with non‑small cell lung cancer (NSCLC) BM, using CA-IX immunostaining. Hypoxia characterization (pimonidazole, CA-IX and HIF-1α) was also performed in different preclinical NSCLC BM models induced either by intracerebral injection of tumor cells (H2030-Br3M, H1915) into the cortex and striatum, or intracardial injection of tumor cells (H2030-Br3M). Additionally, [18F]-FMISO-PET and oxygen-saturation-mapping-MRI (SatO2-MRI) were carried out in the intracerebral BM models to further characterize tumor hypoxia and evaluate the potential of Hypoxia-image-guided-RT (HIGRT). The effect of RT on proliferation of BM ([18F]-FLT-PET), tumor volume and overall survival was determined. We showed that hypoxia is a major yet heterogeneous feature of BM from lung cancer both preclinically and clinically. HIGRT, based on hypoxia heterogeneity observed between cortical and striatal metastases in the intracerebrally induced models, showed significant potential for tumor control and animal survival. These results collectively highlight hypoxia as a hallmark of BM from lung cancer and the value of HIGRT in better controlling tumor growth.

Characterization of hypoxia in murine BM models from lung cancer. To investigate the hypoxic feature of BM and its potential spatial heterogeneity, metastatic adenocarcinoma H2030-Br3M or H1915 cancer cells were injected into the brain of nude rats in two different regions, striatum and cortex. As hypoxia can be caused by low vascularization RECA-1 staining was firstly analyzed and showed a significant decrease in vessel density and an increase in vessel diameter and domains for both cortical and striatal metastases in comparison to healthy tissue for both H2030-Br3M and H1915 models (Fig. 3a). Interestingly, inter-BM heterogeneity was observed in the H1915 model with less vascularization in cortical BM in comparison to the striatal BM (p < 0.05, Table 1). Moreover, intra-BM heterogeneity was observed with a greater vascularization evident in the shell in comparison to the core of the tumor for both H2030-Br3M and H1915 models and both tumor locations. As a consequence of the low vascularization, hypoxia was expected to be strongly present. Indeed, pimonidazole, CA-IX and HIF-1α staining revealed marked staining in both models and regions (Fig. 3b,c). Interestingly, as shown in Fig. 3b, CA-IX and HIF-1α staining was more pronounced in cortical metastases than striatal metastasis in the H1915 model.
The hypoxia feature of BM was further confirmed through additional studies in a BM model induced by intracardial injection of H2030-Br3M cells in nude rats. This model is complementary to the intracerebral model of BM since it mimics the natural route of brain parenchyma invasion. Sixty percent of the animals presented with cortical BM and twenty percent with cerebellar BM. All cortical and one of two cerebellar BM were positive for pimonidazole staining (86% of positives pimonidazole BM). These results are in line with the clinical results concerning the hypoxic status in BM from lung cancers (Fig. 4).
The microenvironment of BM was further studied using multimodal imaging, an approach more adapted for future hypoxia-guided RT. Representative anatomical MRI for the H2030-Br3M and H1915 intracerebral models 20 days after inoculation are shown in Fig. 5a (top). As a related parameter of hypoxia, we firstly analyzed BM vascularization with MRI (CBV-MRI). For both models, MRI revealed similar CBV in cortical and striatal metastases as in healthy tissue. More interestingly, [ 18 F]-FMISO-PET revealed pronounced hypoxia in the H2030-Br3M model in both regions (p < 0.05 and p < 0.001 vs. Control for striatal and cortical metastases and p < 0.05 between the two BM, Fig. 5b). For H1915, the heterogeneity in hypoxia observed by immunohistochemistry between the brain regions was confirmed by [ 18 F]-FMISO-PET. The cortical metastases was more hypoxic than in the healthy tissue and also than in the striatal metastasis, which was not significantly different to healthy tissue (p < 0.001 vs. Control and p < 0.01 vs. striatal metastasis). SatO 2-MRI was also used as another imaging readout of tumor hypoxia. SatO 2 -MRI revealed the presence of hypoxia and its spatial heterogeneity in the H1915 model ( Fig. 5b; p < 0.001 and p < 0.05 for cortical metastases vs. Control and striatal metastasis, respectively).
These results showed, in BM preclinical models from human lung adenocarcinoma, the existence of hypoxia in BM microenvironment. These results support the concept that adapting treatment, especially RT, to hypoxia could improve tumor control. However, taking account of the marked intra-BM and inter-BM heterogeneity, robust tools are needed to map hypoxia. To this end, we have shown that [ 18 F]-FMISO-PET and SatO 2 -MRI are of interest to characterize BM hypoxia and the inter-BM heterogeneity, and consequently may be useful for HIGRT.
Added value of hypoxia image-guided radiotherapy (HIGRT) in preclinical models of BM. The aim of this study was to evaluate the therapeutic potential of quantitatively adapting RT to hypoxia based on imaging for BM. As the intracerebral H1915 model presented the highest inter-BM heterogeneity in terms of hypoxia, the therapeutic potential of HIGRT was evaluated in this model. To address the quantitative boost of RT needed to counteract hypoxia-induced radioresistance, Oxygen Enhancement Ratio (OER) was calculated in vitro in this cell line. Clonogenic assay performed under normoxia and hypoxia (1% of O 2 ) revealed an increase in survival fraction at 2 Gy (SF2) in hypoxia leading to an OER of 1.33 (p < 0.05, Fig. 6a). To further support the concept that radioresistance due to hypoxia is a hallmark of BM, a clonogenic assay was also performed on the H2030-Br3M cell line and similar results were obtained (Fig. 6b). As presented in Fig. 6c, SatO 2 -MRI was used to define a central hypoxic region to guide RT boost by a factor of 1.33 in hypoxic regions depending on the OER calculated in vitro. More details are presented in the "Methods" section.   Fig. 7a-c, this treatment was not able to control the hypoxic cortical BM, and a recurrence occurred only in these tumors (Fig. 7e). In contrast, in the non-hypoxic striatal metastasis, RT controlled tumor growth more effectively (Fig. 7d). The therapeutic effect of RT resulted in limited overall survival, although significantly greater than the Control group (Fig. 7f). HIGRT controlled both non-hypoxic and hypoxic BM and was supported by a significant decrease in cortical metastasis proliferation in comparison to Control and RT groups (p < 0.01 and p < 0.05 vs. Control and RT groups, respectively, Fig. 7c). Ultimately, HIGRT was able to control definitively three of the five (60%) non-hypoxic and hypoxic BM, with a significant increase in overall survival (p < 0.01, Fig. 7f

Discussion
Current treatment options for patients with BM from lung cancer remains limited, firstly because BM are detected too late and standard external beam RT shows response heterogeneity between patients 21 . Secondly, RT may induce a decline in cognitive ability for long-term survivors 5 . Here we hypothesize that treatment's response heterogeneity could be due to both inter-patient and inter-metastasis heterogeneity notably in terms of the microenvironment and, in particular, hypoxia 22 . Precise characterization of the BM microenvironment, including hypoxia could enable more personalized RT and, consequently, better tumor control.
In this study, we first investigated whether hypoxia is present in the microenvironment of BM from primary lung cancer at both clinical and preclinical levels. In the patient's biopsies from non-small cell lung cancer (NSCLC) BM, CA-IX staining was used as an endogenous molecular marker of hypoxia as it is the most used in Table 1. Quantification of the RECA-1 immunostaining. Data represent the mean (SD) of vessels density and diameter and domains analyzed with histology for H2030-Br3M and H1915. Mean ± SD, n = 3, *p < 0.05, **p < 0.01 and ***p < 0.001 versus healthy tissue, # p < 0.05 and ### p < 0.001 versus cortical metastasis core and $ p < 0.05 versus striatal metastasis core.  23 . In patient studies, Proescholdt and colleagues have shown CA-IX expression in BM from lung, breast, kidney and melanoma cancer, and matched expression of CA-IX between primary lung cancer and BM was also found in another clinical study 24,25 . We have also further shown in the present manuscript, from four biopsies selected for their high or low CA-IX staining, concordance between both CA-IX and HIF-1α staining. Hypoxia was further characterized in different preclinical models of BM from lung cancers (H2030-Br3M, H1915; via intracerebral and intracardiac injections) using immunohistological staining (pimonidazole, HIF-1α) as well as imaging approaches. Subsequently, we assessed whether a therapeutic approach based on hypoxia imaging, HIGRT, has potential to counteract intra-and inter-BM heterogeneity. To reflect the clinical situation as closely as possible and to mimic the multiple foci observed in the clinical situation, we used preclinical models of BM in which human metastatic lung cancer cells were injected into two different brain regions (cortex and striatum).
Other studies have evaluated the presence of hypoxia in BM from lung cancer 8,9,26 . For example, Berghoff and colleagues have also shown, with HIF-1α staining, that hypoxia is a prognostic factor for overall survival 8,9 . These results highlight the importance of adapting RT according to the hypoxic status. Previous studies have shown the presence of hypoxia in BM preclinical models from primary breast cancer 27,28 and none in preclinical models of BM from lung cancer. In this study, both intracerebral and intracardiac BM models of lung cancer showed hypoxic feature of BM. Of note, we have also obtained similar results from a breast cancer BM model induced by intracardial injection of human breast cancer cells (MDA-231 cell line) in nude mice. Indeed, two hundred thirty five BM from seven mice were analyzed and reported 69% positive pimonidazole staining (representative images of immunostained sections are presented in Sup Fig. 3; Sup Material).
Moreover, the preclinical existence of inter-BM microenvironment heterogeneity, in terms of hypoxia, according to the location (cortex vs. striatum) has not previously been reported. Here, we showed that the same number of cells injected at the same time in two different brain regions gave rise to tumors with similar BM volumes but different degrees of hypoxia in the H1915 and H2030-Br3M models. Interestingly, a preclinical study from www.nature.com/scientificreports/ Perera and colleagues showed, with intracardiac injection models, that BM preferentially developed in the cortex in comparison to deep brain 29 . Furthermore, two clinical studies have shown that BM more frequently occur in cortical brain than deep brain structures 30,31 . These data suggest that difference exist in the microenvironment of www.nature.com/scientificreports/ healthy cortex with other brain regions which could lead to an easier BM development in the cortex. Interestingly, we obtained preliminary results showing that acetyl-CoA content, a central metabolite 32 , is present at higher concentrations in healthy cortex compared to healthy striatum, a difference that is maintained in the presence of BM (Sup Fig. 4, Sup Material). These differences in metabolism/nutrients between the brain regions may differentially influence cancer cells, for example in term of proliferative rate, as underlined by the literature [33][34][35][36] and also our results obtained by [ 18 F]-FLT PET analyses (Sup Fig. 5, Sup Material). Indeed, the acetyl-CoA pool is the key metabolite to meet metabolic rewiring and the high bioenergetic demands of BM 37,38 . In addition, a clinical study has shown that acetyl-CoA metabolism is important with high level of acetyl-CoA oxidase 1 in BM from breast cancer in comparison to other metastatic sites, however differences between BM themselves was not investigated 39 . The present work, together with those previously published 24,26,27,[40][41][42][43][44][45][46][47][48][49] , highlight the importance of (i) characterizing the BM microenvironment in individual patients, and (ii) adapting external RT to the presence of hypoxia for better tumor control. Zakaria and colleagues have highlighted the importance of using more functional MRI biomarkers, such as diffusion-MRI, rather than conventional contrast-enhanced MRI for RT planning to improve BM control 50 . On the basis of our initial findings, we proposed that hypoxia imaging could enable external RT to be adapted to the tumor hypoxic environment (HIGRT). In this study, we showed that both [ 18 F]-FMISO-PET and SatO 2 -MRI can detect hypoxia heterogeneity between cortical and striatal BM. SatO 2 -MRI for HIGRT, rather than [ 18 F]-FMISO-PET, owing to the higher spatial resolution of MRI compared to PET, enable intra-metastasis hypoxia heterogeneity to be determined and guide precise HIGRT. In addition, the RT boost was based on OER calculation on in vitro experiments and was applied when the SatO 2 -MRI fell below a threshold from the literature (40%) 51 , resulting in a binary application of HIGRT. Other approaches have suggested to use non-linearity correlation between hypoxia imaging and real oxygen pressure in tissue (ptO 2 ) for dose modulation in primary brain tumors 52,53 , however this approach has still to be validated at the preclinical and clinical levels. Integration into the clinic setting of functional imaging, like hypoxia imaging, in the RT workflow is still challenging. As we discussed in a previous review 54 , MRI based oxygenation biomarkers, such as SatO2 or Oxygen enhancement-MRI, would have the advantage of achieving high resolution in comparison to PET biomarkers, such as [ 18 ]-FMISO or [ 18 ]-FAZA. However, such MRI-based approaches would have to be considered as an indirect assessment of oxygen level, as they depend on the vascular compartment, whereas PET biomarkers are a direct assessment of cell hypoxia. From these imaging biomarkers, dose calculation can be computed either by contour using thresholds (as performed in this study), or by pixel. The latter is more dependent on the hypoxia imaging resolution, as well as the ability of the RT system to deliver multiple and small beams. In the clinical situation, inverse planning allows calculation of the optimal beams to reach a desired dose. Application of HIGRT for BM depends on whether the clinical center has a stereotactic radiosurgery system, which could enable delivery of a conventional dose in non-hypoxic tumor with a boost applied by dose painting according to contour in hypoxic BM. Dose painting by contour in a hypoxic BM would have the advantage, in comparison to whole tumor dose escalation, of being as close as possible to the hypoxic region and limiting non-useful dose deposition in both non-hypoxic tumor regions and nearby healthy brain tissue. However, as is done with the difference between the CTV and the PTV, a margin obtained from the hypoxia imaging could be added to the biological target volume (BTV) to take into account the uncertainties regarding clinical hypoxia imaging 55 .
In vitro analyses of OER is not possible to be undertaken in the clinic, however, this preclinical study aimed to understand more deeply the interest of HIGRT approach in terms of biological effect.
However, for the main part of this study, only two cell lines from metastatic human lung adenocarcinoma, were used (H1915 and H2030-Br3M) and further preclinical studies would be necessary to determine whether the presence of hypoxia and the therapeutic benefit of HIGRT are conserved in BM originating from other cancers such as breast and melanoma, the other major primary with a propensity to spread to the brain 21 .

Conclusion
We show, at the clinical and preclinical levels that hypoxia is a hallmark of the microenvironment of BM from NSCLC lung cancer. BM heterogeneity in terms of hypoxia between patients and between metastases in preclinical models, highlights the potential interest of personalizing external RT to hypoxia to increase tumor control in comparison to conventional external RT.

Methods
Cell culture. H2030-Br3M adenocarcinomas of human origin that preferentially metastasizes to the brain and H1915 (ATCC, CRL5904) were used for this study. The cell lines were grown in supplemented DMEM (SIGMA-ALDRICH) at 37 °C in wet atmosphere. Rat brain metastases model. All animal investigations were performed under the current European directive (2010/63/EU) including ARRIVE guidelines. This study was undertaken in the housing and laboratories #F14118001 and with the permission of the regional committee on animal ethics (C2EA-54 CENOMEXA, projects #5065-#8941). Nude athymic rats (200-250 g, 8 weeks, female) were maintained in specific pathogen free housing. Rats were manipulated under general anesthesia (5% isoflurane for induction, 2% for maintenance in 70%N 2 O/30%O 2 ). Body temperature was monitored and maintained at 37.5 ± 0.5 °C throughout the experiments. For the BM model, rats were placed in a stereotactic head holder and a scalp incision was performed along the sagittal suture. To investigate potential intermetastases hypoxia heterogeneity, two burr holes of diameter 1 mm was drilled in the skull, 3 and 3.7 mm lateral left and right respectively to the Bregma. H1915 and H2030-Br3M cells (5 × 10 4 cells in 3 μl-PBS containing glutamine 2 mM) were injected over 6 min via a fine needle (30G) connected to a Hamilton syringe. The injection sites were the left caudate putamen at a depth of www.nature.com/scientificreports/ 6 mm and the right cortex at a depth of 2.5 mm. Animals were then followed by anatomical MRI over a 21 days period to follow BM development, then animals follow one of the two substudies detailed in Fig. 1a,b. Twentyone days period between injection and treatment/imaging was chosen to allow measurable hypoxia with PET imaging owed to its spatial resolution. Methods for the intracardiac model of BM is detailed in Sup Material).

Study 1: Characterization study of hypoxia in BM.
Patients. Between May-2014 and May-2019, the brain tumor registry identified 28 patients aged over 18 years with a diagnosis of BM from primary NSCLC lung cancer. Overall, 28 patients (9 women/19 men) with a median age of 63.5 years were retrieved with sufficient tissue available for additional biomarker studies. All tumor specimens were reviewed by an experienced neuropathologist (FC): the pulmonary origin of BM was certified by the presence of a positive immunostaining for thyroid transcription factor-1 and cytokeratin 7 within the BM. As required by French laws, all patients provided informed consent, and the study was approved by the institutional ethics committee (North-West-Committeefor-Persons-Protection-III N°DC-2008-588). All methods involving humans were carried out in accordance with local guidelines and regulations and with the Declaration of Helsinki.

Immunohistochemical characterization of hypoxia in BM biopsies. Hypoxia in BM biopsies
was evaluated through the carbonic anhydrase-IX (CA-IX) expression, a recognized endogenous markers of hypoxia 20 and HIF-1α. Slides were incubated with primary antibody against CA-IX (EPITOMICS, 1/100-30 min room temperature) and HIF-1α (Cell Signaling #36,169, 1/100-overnight at 4 °C) and revealed using the Novolink (LEICA) kit 56 . All slides were examined by one expert pathologist and positives or negatives CA-IX-BM were reported, which allows to obtain percentage of positives CA-IX-BM.

Immunohistochemical characterization of hypoxia and vascularization in BM preclinical models.
At the time of the last oxygenation and vascular MRI session (D22), immunohistochemical staining for rat-endothelial-cell-antigen (RECA-1), pimonidazole, CA-IX and HIF-1α were used as previously described 14 to characterize vascularization and hypoxia, respectively. For the RT therapy study (study 2), 3 days after initiation of treatments (D24), immunohistochemistry was performed to assess tumor cell proliferation using Ki67.
Positron emission tomography (PET). [  www.nature.com/scientificreports/ (reference) and the X-ray scan (input) by normalized mutual information. Whenever necessary, the registration was manually refined. As a second step, PET parameters were co-registered to MRI parameters. Clonogenic survival assay. To characterize the radioresistance due to hypoxia in H1915 cells, a clonogenic assay was used in normoxic and hypoxic conditions 60 . For normoxic condition, BM cells were plated in 6-well plates (1500 cells/well), exposed to X-rays (0 Gy; 2 Gy; 4 Gy; 6 Gy and 8 Gy) 24 h after the cell seeding and kept in the incubator for a duration of 12 days. Then, colonies were stained with 2% crystal violet (Sigma-Aldrich) diluted in 20% ethanol. For hypoxic condition, cells were plated in 6-well plates (1500 cells/well) kept in an hypoxic chamber for 6 h (1% O 2 /5% CO 2 , 37 °C, Invivo2-500, RUSKINN, ABE) 18 h after cell seeding and then exposed to X-rays before keeping in the hypoxic chamber for 12 days. A value of 1% O2 was chosen as it corresponds to a partial pressure of oxygen of 6.6 mmHg, which is close to previous reports of pO2 in brain tumors and lung cancer 53,61,62 . The effect of oxygen level in radiation efficacy was evaluate with quadruplicate experiment using, for each experiment, four wells per radiation dose. SF2, (survival fraction at 2 Gy), D37 (lethal dose corresponding to dose for SF = 37%) and OER (oxygen enhancement ratio) were obtained from the survival curves as previously described 63 . In vivo radiation treatments. After SatO 2 -MRI imaging session, at day 21, rats (n = 7 per group) were assigned in the three following treatment groups: Control, Whole Brain Radiation Therapy (WBRT) and Hypoxia Image-Guided Radiotherapy (HIGRT). These RT treatments were performed using a XRad-225Cx irradiator, (225 kV X-rays; 3.3 Gy/min, Precision X-Ray, Equipex Rec-Hadron, CYCERON biomedical imaging platform, Caen). A cone beam CT (CBCT) image was acquired using the XRad-225Cx irradiator for anatomical delimitation. Treatment planning and subsequent beam delivery were performed using SmART-Plan software (Precision X-Ray, USA and MAASTRO Clinic, Netherlands), with segmentation, targeting and planning performed using the CBCT image. For RT treatment, animals received panencephalic RT with a total dose of 12 Gy in 3 fractions of 4 Gy, using a 15 mm collimator under CBCT guidance. For HIGRT, SatO2-MRI was used to identify non-hypoxic and hypoxic tumor areas. Areas were considered hypoxic if the SatO2 level was below 40% 51 . SatO2-MRI was co-registered to the CBCT image from the irradiator, which allowed a co-registration matrix to be obtained and this was used to transfer non-hypoxic and hypoxic tumor localizations into the irradiator system geometry. Striatal normoxic BM received a 12 Gy dose using a 5 mm collimator. Hypoxic cortical metastases received 12 Gy with the 5 mm collimator, plus an additional dose of 4 Gy in the hypoxic area using a 2.5 mm collimator, to achieve a total dose of 12 Gy × OER. Based on the clonogenic assay, the OER was 1.33, which leads to a total dose in the hypoxic tumor area of 16 Gy (Fig. 6c).
Survival study. After irradiation treatment, rats were followed for a survival study. Prior to the initiation of the study, we defined 80 days as an arbitrary endpoint for overall survival outcome or when total tumor volume overcome 150 mm 3 or when tumor impair mobility.
Statistical analyzes. All data are presented as mean ± SD. Student's t-test was used to compare cortical and striatal metastases and one-way and two-way (group and time effects) ANOVA followed by Tukey's post-hoc test was used to compare differences between treatment groups. A log-rank test was used to compare Kaplan-Meier curves. Statistical analyzes were obtained with JMP (SAS-Institute-Inc, USA).
Ethics approval and consent to participate. As required by French laws, all patients provided informed consent, and the study was approved by the institutional ethics committee of Caen University Hospital (North-West Committee for Persons Protection III CPP N ° DC-2008-588 of Caen University Hospital, France), France. For animal experiment: all applicable international, national, and/or institutional guidelines (including the ARRIVE's guidelines) for the care and use of animals were followed. All procedures performed in studies involving animals were in accordance with the ethical standards of the institution or practice at which the studies were conducted as detailed in the "Methods" section.

Data availability
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.